ci(pr-test): use GitHub API to get changed files + diff

[yin1999]

Description

get diff file and changed files' name from GitHub API is much faster.

Motivation

as gh is an built-in GitHub cli in Action runner, we can use this tool to get changed files, which is much faster than git diff command (must fetch the history commits).

Additional details

using gh cli

Only used 0 seconds.

https://github.com/yin1999/translated-content/actions/runs/3580409184/jobs/6022509363#step:6:4

using git diff (used by technote-space/get-diff-action)

used 40 seconds.

https://github.com/mdn/translated-content/actions/runs/3163384214/jobs/5150888786#step:7:1

conclusion

Using gh cli could speed up the PR test workflow. But when we use this, we must get the diff file from https://github.com/ower/repo/compare/{base}...{head}.diff and this diff file just show the diff in changed files (which may differ from git diff which will collected the changes in the main branch).

[yin1999]

If this is ok, I'd like to reflect the changes in mdn/content

[caugner]

LGTM and promising, thanks! 🎉

Just some comments:

[yin1999]

My bad, I've ignored to export absolute path here, see testing pr (yin1999#8):

• failed with image (unused): https://github.com/yin1999/translated-content/actions/runs/3293923349
• passed with image (used): https://github.com/yin1999/translated-content/actions/runs/3293947316/jobs/5430964574

Now, this seems ok.

[yin1999]

marked as draft until I figure out the workflow run (https://github.com/yin1999/translated-content/actions/runs/3293826528/jobs/5430706523). It seems to be not right.

[yin1999]

There is a test I've made before

https://github.com/yin1999/translated-content/actions/runs/3580409184/jobs/6022509363#step:8:6

[yin1999]

For there maybe too many files has been changed in a single PR, the gh pr view command used before can only fetch up to 100 changed files. But for almost all the PR, the commit in it would be less than 30 commits. So we can use github rest API to get changed files. (See https://docs.github.com/rest/commits/commits?apiVersion=2022-11-28#compare-two-commits). This API will limit 30 commits per page. So for those must few PRs, we can use --paginate (see options in https://cli.github.com/manual/gh_api) to fetch all the commits.

Addition: The API limitation from GitHub Actions with GitHub Token would be 1000 requests per hour. for a single repo This should be enough for most of cases.

[yin1999]

We can check this PR as an example: #9513

PR info: https://api.github.com/repos/mdn/translated-content/pulls/9513
Compare info (all the 183 files in in the single rest response, see field files): https://api.github.com/repos/mdn/translated-content/compare/a9df62bb4885e5d9cdccd3834d5a24c921ef7cfa...16f5fc0db4ab1f5cb3858ab7878c428c19e9a79e

[yin1999]

If this is ok. Should we try it in Markdown lint (PR) first? There are some failed workflows (cannot get changed files), and it is less important than the current CI.

[yin1999]

We can see the diff url in https://docs.github.com/rest/commits/commits#compare-two-commits

This may avoid an additional api call (making it harder to reach the limitation)

[caugner]

@yin1999 This has probably been discussed before, but why can we actually not use git ls-files? My expectation would be that the workflow run in the context of the PR has the history of all commits between BASE_SHA and HEAD_SHA. Is this not so?

[yin1999]

My expectation would be that the workflow run in the context of the PR has the history of all commits between BASE_SHA and HEAD_SHA. Is this not so?

Actually not (see: actions/checkout#520), the actions/checkout@v3 action could only fetch specified n depth (with fetch-depth input). By default, the param would be 1 (equivalent to git clone --depth=1), so the commit history between BASE_SHA to HEAD_SHA will not be included in the workflow context.

If we need to fetch commits that are in a single PR (branch), we need to use git fetch with enough depth (technote-space/get-diff-action's implement) which may take too much time.

And I've created another PR for markdown-lint (PR), to deal with the workflow failure. If the solution seems ok, could we try on markdown-lint. Because this workflow may be more important for us.

[caugner]

@yin1999 Understood, thanks for clarifying!

Could you update this PR to align with the changes in #10390 (i.e. without the temporary file)? And then it should be good to go. 🙏

[yin1999]

Done in 6faf7b6. I've tested this in my repo, see: https://github.com/yin1999/translated-content/actions/runs/3630604417/jobs/6124154109

[caugner]

Thank you @yin1999, this is awesome! 🎉